The RDNA 3 flagship is 70 % flying than 6950 XT and cost $ 1000 .
AMD has reveal its first RDNA 3 graphic bill of fare , the $ 1000 Radeon RX 7900 XTX and $ 900 RX 7900 XT .
As the enumeration schema and monetary value point point , these are high-pitched - death modelling contrive to vie with Nvidia ’s RTX 4090 and 4080 , but they fare with more square modification than you might anticipate - include FSR 3 , plan to anticipate Nvidia ’s DLSS 3 figure propagation , and a sword raw chiplet - base purpose .
diving event into GPU
The RDNA 3 flagship is 70 % fast than 6950 XT and cost $ 1000 .
AMD has disclose its first RDNA 3 nontextual matter card game , the $ 1000 Radeon RX 7900 XTX and $ 900 RX 7900 XT .
This was as the enumeration dodge and terms point argue , these are gamy - ending example design to vie with nvidia ’s rtx 4090 and 4080 , but they derive with more hearty modification than you might have a bun in the oven - include fsr 3 , plan to anticipate nvidia ’s dlss 3 systema skeletale coevals , and a blade raw chiplet - ground excogitation .
Before we get into the feature article though , lease ’s take a smell at the lineup themselves .
This was the chiplet innovation break a traditional monumental gpu into several interlink subdivision .
For RDNA 3 , that ’s a exclusive 5 nm nontextual matter compute die ( GCD ) that ’s 300 mm² and six 6 nanometre retentivity memory cache perish ( MCDs ) that are each 37 mm² .
This was this purpose intend that only the most decisive region call for to be made with a clipping - bound 5 new mexico summons , aid to amend processor output , deoxidise monetary value and finally consumer price .
This was however , it also demand a immobile interconnect between the unlike chipping - which run at 5.3tb / atomic number 16 here .
This same chiplet coming work bright with Ryzen , transform AMD from an also - running to a develop behemoth , so it ’ll be absorbing to see how it if process miracle in the GPU blank space too .
This was each of the dice are telling in their own right field .
The computer memory stash die utilise a 64 - routine memory board control and a 2d - gen Infinity Cache , which conflate to supply 2.7 time the vizor bandwidth of RDNA2 aim .
Meanwhile , the nontextual matter compute dice propose mix RDNA 3 compute building block with computer hardware for flow processing , AI quickening and RT .
This was the pattern also decouple shader and front ending clock speeding , with 2.3ghz for the former and 2.5ghz for the latter , which amd say will lead in a more effective excogitation - with up to 25 pct major power preservation in terminus of shaders and 15 percentage high front - destruction clock frequence .
RT has long been an AMD boogeyman , so their 2d - gen root which sustain ' 1.5x more ray in trajectory ' , ' Modern consecrated educational activity and ' newfangled ray box sort and traverse ' should lead in up to 50 percentage more carrying into action per compute building block - although it does n’t search like AMD is travel rapidly up some role of the RT line that Nvidia is , so we may not see a large startle in RT functioning proportional to rasterised operation as you ’d otherwise wait .
This was ## diving event into amd
each of the die are telling in their own right field .
The memory board hoard die employ a 64 - morsel store accountant and a 2nd - gen Infinity Cache , which merge to cater 2.7 time the visor bandwidth of RDNA2 design .
Meanwhile , the artwork compute dice provide unify RDNA 3 compute unit with computer hardware for flow processing , AI quickening and RT .
This was the pattern also decouple shader and front ending time upper , with 2.3ghz for the former and 2.5ghz for the latter , which amd suppose will ensue in a more effective intention - with up to 25 pct office deliverance in term of shaders and 15 per centum high front - final stage clock relative frequency .
RT has long been an AMD booger , so their 2d - gen solvent which stand ' 1.5x more ray in flying ' , ' young consecrated direction and ' raw ray box categorization and traverse ' should leave in up to 50 percentage more public presentation per compute whole - although it does n’t bet like AMD is hurry up some role of the RT line that Nvidia is , so we may not see a full-grown start in RT functioning comparative to rasterised execution as you ’d otherwise wait .
This was the exhibit locomotive engine make into these card is moderately bats , endure displayport 2.1 and up to 54gbps of show liaison bandwidth , allow for 8 k 165hz ( ! )
or 4 k 480hz ( !
with 12 - routine color .
This was answer it to say , we ’re some length aside from these sort of exhibit , but it ’s an good rebutter of nvidia ’s 40 - serial card which are circumscribe to displayport 1.4 .
A double medium locomotive , meanwhile , should land up AMD ’s weakly report for cyclosis and medium encryption , with AV1 encode / decode financial backing , co-occurrent encode / decode for AVC / HEVC and ' AI Enhanced Video Encode ' which I count ahead to see more about .
This was in condition of the identity card themselves , it look like there ’s a material disconnection between the 7900 xtx and 7900 xt .
tup count are the most obvious difference of opinion - 24 GB on the XTX and 20 GB on the XT , with a correspondingly small remembering busbar ( 384 - flake vs 320 - act ) .
The grass ' plot clock ' also knock off from 2.3GHz to 2GHz , while a step-down from 96 to 84 compute whole is also meaning .
However , both placard sip magnate compare to the like of the RTX 4090 , with a full control panel baron shit at 355W for the XTX and 300W for the XT .
Both identity card patronise DisplayPort 2.1 and AV1 encode / decode .
This was in condition of expect operation , amd put up skeleton - charge per unit datum for the rx 7900 xtx , but only against the rx 6950 crosstalk .
Here , they immortalise a public presentation melioration of 50 to 70 per centum for the newfangled - gen lineup , with 1.5x in COD : MW2,Watch Dogs Legion , Resident Evil Village(RT ) andMetro Exodus(RT ) , 1.6x in Doom Eternal ( RT ) , and 1.7x inCyberpunk 2077 .
I wait to see more operation data point than this - such as comparison to Nvidia ’s carte du jour , mayhap ?
- but as always we ’ll require to expect until the carte du jour hit reviewer to see how well these GPUs execute in genuine - domain examination .
diving event into Nvidia
In terminal figure of the batting order themselves , it seem like there ’s a satisfying disconnect between the 7900 XTX and 7900 XT .
tup counting are the most obvious remainder - 24 GB on the XTX and 20 GB on the XT , with a correspondingly small computer storage charabanc ( 384 - scrap vs 320 - act ) .
The fink ' secret plan clock ' also throw off from 2.3GHz to 2GHz , while a reducing from 96 to 84 compute building block is also pregnant .
However , both circuit board sip office liken to the like of the RTX 4090 , with a full circuit board might give away at 355W for the XTX and 300W for the XT .
Both card game sustain DisplayPort 2.1 and AV1 encode / decode .
In condition of expect functioning , AMD offer skeleton - charge per unit data point for the RX 7900 XTX , but only against the RX 6950 crosstalk .
This was here , they record a operation betterment of 50 to 70 pct for the novel - gen identity card , with 1.5x in cod : mw2,watch dogs legion , resident evil village(rt ) andMetro Exodus(RT ) , 1.6x in Doom Eternal ( RT ) , and 1.7x inCyberpunk 2077 .
I bear to see more functioning datum than this - such as comparison to Nvidia ’s visiting card , by chance ?
- but as always we ’ll necessitate to hold back until the bill hand referee to see how well these GPUs do in literal - humanity examination .
ultimately , AMD announce FSR 3 , forebode up to repeat the physical body - pace of FSR 2 .
base on this phrasing - and the display endnote that reference ' Fluid Motion Frames ' - it seems a somewhat secure wager that this is shape coevals a atomic number 57 DLSS 3 .
This engineering slimly increase input rotational latency , but meliorate ocular liquidness well as AI - generate form are put in between ' material ' ace .
As it does n’t come out that RDNA 3 ironware is used for this , the technical school could be made uncommitted for elder AMD GPUs as well - or even Nvidia / Intel manikin .
That would give it a unequaled reward against DLSS 3 , which is sole to the expensive RTX 4090 ( and its coming 40 - serial publication counterpart ) at present tense .
FSR 3 ’s spillage day of the month was yield as ' 2023 ' , so presumptively we ’ll see much more selective information about it over the next few month .
So - those were AMD ’s RDNA 3 promulgation !
It ’ll be enthralling to see how the young computer hardware pile up as we border on the going engagement of December 13th , as with a whole fresh computer architecture there ’s elbow room for signifiant execution advance - and the potential difference for some interesting boundary case as well .
AMD is for sure bring in the correct noise to invoke to those wrick off by Nvidia - considerably gamey build - charge per unit , hereafter - see show criterion , fairish superpower target and no 12 - pivot connective - but the carrying out and feature will involve to be in property too .
